HIGHWAY TO HITLER Nico Voigtlaender Hans-Joachim Voth Working Paper 20150
http://www.nber.org/papers/w20150
NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue
Cambridge, MA 02138 May 2014, Revised January 2021
For helpful comments, we thank Sascha Becker, Eli Berman, Tim Besley, Leonardo Bursztyn, Davide Cantoni, Bruno Caprettini, Ernesto Dal Bo, Melissa Dell, Rick Hornbeck, Torsten Persson, James Robinson, Andrei Shleifer, Jim Snyder, and Noam Yuchtman. Seminar audiences at Basel University, Bonn, CREI, Dartmouth, Haas-Berkeley, King’s College London, the Juan March Institute, LSE, UC-Berkeley, UC San Diego, Warwick, Yale, Zurich, and at the Barcelona Summer Forum offered useful suggestions. We are grateful to Hans-Christian Boy, Vicky Fouka, Cathrin Mohr, Casey Petroff, Colin Spear and Inken Töwe for outstanding research assistance.
Maja Adena, Ruben Enikolopov, and Maria Petrova kindly shared data on radio signal strength in Nazi Germany. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research.
NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications.
© 2014 by Nico Voigtlaender and Hans-Joachim Voth. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.
May 2014, Revised January 2021 JEL No. H54,N44,N94,P16
ABSTRACT
Can autocracies win electoral support by showcasing economic competence? We analyze a famous case – the building of the Autobahn network in Nazi Germany. Using newly collected data, we show that highway construction was effective in boosting popular support, helping to entrench the Nazi dictatorship. Direct economic benefits such as declining unemployment near construction sites are unlikely to explain the increase in pro-Nazi votes. In addition, Nazi propaganda used the Autobahn as a powerful symbol of successful economic policy, putting an effective end to austerity – so that many Germans credited the Nazi regime for the economic recovery. In line with this interpretation, we show that support for the Nazis increased even more where highway construction coincided with greater radio availability – a major source of propaganda. The effect of highways was also significantly stronger in politically unstable states of the Weimar Republic. Our results suggest that infrastructure spending can raise support for autocracy when voters are led to associate it with visible economic progress and an end to political instability.
Nico Voigtlaender
Anderson School of Management University of California, Los Angeles 110 Westwood Plaza
C513 Entrepreneurs Hall Los Angeles, CA 90095 and NBER
nico.v@anderson.ucla.edu Hans-Joachim Voth Department of Economics University of Zurich Schoeneberggasse 1 CH-8001, Zurich Switzerland
joachimvoth@gmail.com
A online appendix is available at http://www.nber.org/data-appendix/w20150
“…while a decade of Weimar parliaments had produced only talk and sketches, a mere three years of National Socialism had built a thousand kilometres of traversable superhighways… Their very existence seemed to verify the Nazi thesis that the state must be given a free hand, if it were to restore Germany to her former glory.”
(Shand 1984, p.194).
1 Introduction
In the last 200 years, democracy has spread around the globe. In 1816, less than 1% of world population lived in democracies. In 2015, the figure stood at 58%. However, progress has not been a one-way street: Between 1920 and 1938, the number of democratic states in the world fell from 20 to 13, and the share of the population living in democracies declined from 20.5%
to 14.8%.1 Nor is democratic fragility necessarily an issue of the past: Ever fewer young Europeans and U.S. citizens consider it essential to live in a democracy. In the U.S., their share declined from more than 70% among the 1930s birth cohort to about 30% for the 1980s cohort.2
For democracy to fail, autocracy has to triumph. Authoritarian leaders rarely rule by force alone, and many rely on popular support (Egorov and Sonin 2014). How do authoritarian rulers convince voters to discard and disregard their democratic rights? The decline of democracy is often associated with (perceived) social disorder. In these circumstances, dictatorships may look appealing to the masses because of their capacity to restore order (Finer 2002; Djankov et al. 2003). As Easterly and Pennings (2016) noted, the view that economic growth is easier to engineer under strong autocratic regimes has recently gained in prominence, with China and Singapore often cited as leading examples.3 Similarly, thirty percent of US respondents in the 2010 World Value Survey thought that it would be “good or very good to have a strong leader” that does not have to “bother with parliaments and elections”
(Foa and Mounk 2016). But do shows of efficacy, indeed, buy support for budding dictators?
1 Boix et al. (2012), Vanhanen (2010). Along the same lines, Birdsall and Fukuyama (2011) observe that
“political leaders in the developing world now associate efficiency and capability with autocratic political systems.”
2 Foa and Mounk (2016). The question asked is whether respondents feel “it is essential to live in a country that is governed democratically,” where “feeling strongly” corresponds to a rating of 10 on a 10-point scale.
3 See also Friedman (2009). Jones and Olken (2005) show that turnover in the leadership of autocracies leads to sharp changes in economic performance; Besley and Kudamatsu (2008) model the institutional features of autocracies that can increase their chances of success.
It may instead be true that entrenched dictators are more effective, rather than effective dictators becoming entrenched.
In this paper, we aim to identify the causal effect of effective policy implementation on autocratic consolidation. We study a concerted effort showcasing an authoritarian state’s ability to accomplish ambitious goals: the construction of the Autobahn in Nazi Germany, the world’s first high-speed road network. Initially, the Nazi grip on power was far from absolute.
In an election that was neither free nor fair in November 1933, many voters opposed the Nazi regime – in several major cities, more than a quarter of votes were cast against the Nazi Party.4 By mid-1934, the regime’s popularity was waning amongst the middle class; conservative elites were dismayed by Nazi lawlessness, and President von Hindenburg threatened military rule.5 And yet, by the late 1930s, the Nazi regime had become one of the most popular in German history – a “consensual dictatorship.”6 As late as 1955, almost half of all Germans opined that Hitler would have been “one of the greatest German statesmen had it not been for the war” (Möbius 2013: 257).7
A key turning point in the regime’s fortunes came in August 1934, after President von Hindenburg’s death. Hitler became both chancellor and president, concentrating singular authority in the hands of the Führer.8 This increase in power was overwhelmingly endorsed in a popular referendum. We argue that Autobahn construction contributed markedly to the regime’s popularity, as reflected in major support for the referendum.9 Since the last election, held just 10 months before, road construction had begun in earnest. Opposition against the regime declined significantly where the new roads were being built. Figure 1 illustrates our
4 Voting results from the Nazi period cannot be taken at face value. Intimidation was massive, and there is some (limited) evidence of fraud. Nonetheless, we argue that information about popular support can be extracted from vote shares. Yes votes did not necessarily reflect genuine support – but no-votes were a clear sign of opposition.
No-votes varied importantly over time and space. Even large cities recorded substantial differences: In Aachen in 1934, for example, 24% voted “no”; in Nuremberg, on the other hand, only 4.6% voted against Hitler becoming both chancellor and president
5Time Magazine in July 1934 quoted Adolf Hitler as saying: “Don't forget how people laughed at me 15 years ago when I declared that one day I would govern Germany. They laugh now, just as foolishly, when I declare that I shall remain in power!”
6 Aly (2005) and Bajohr (2005).
7 In another survey, some 10% opined that even with the war, Hitler was the greatest German statesman of all time, whose eminence would only be accepted in the future (Noelle and Neumann 1956: 135).
8 In addition to the referendum, the wholesale murder of the SA-leadership and other prominent anti-Nazis in the Night of the Long Knives allowed Hitler to consolidate his powers.
9 Saiz (2006) notes a close association between dictatorships and highways, but argues that this is explained by their use as instruments of repression.
main finding. It shows the change in support for the Nazi regime by distance to Autobahn construction.10 The Nazis gained more support the closer locations were to highway construction. Our results suggest persuasion rates of 8-17% due to the Autobahn – a high value compared with other studies (Della Vigna and Gentzkow 2010).11
Motorway planning may have followed a political lead after 1933. To deal with potential endogeneity, we construct least-cost paths between terminal cities connected by highways.
Building costs reflect geological characteristics such as the steepness of the terrain, the number of rivers to be traversed, etc. We use these least-cost paths as an instrument for actual construction, excluding the terminal cities themselves from the analysis. Our IV results confirm the OLS estimates both in terms of magnitude and statistical significance.
What accounts for the Autobahn’s success in winning electoral support for Hitler’s regime?
We argue that highway construction worked because it convinced voters of the Nazi regime’s
“competence,” mainly by seeming to solve Germany’s key economic problem:
unemployment. Between 1933 and 1934, nationwide unemployment fell by half (Humann 2011). However, the roads’ contribution to this success was limited; their construction employed relatively few workers, and a cyclical upswing had already begun to reduce unemployment before road-building commenced (Ritschl 1998). Correspondingly, we find that unemployment fell at a similar rate in cities near highway construction and those further away. Nevertheless, Goebbels’ propaganda effectively sold the notion that Nazi roadbuilding was reviving the German economy.12 Thus, news that the highway was coming was enough to sway voters: We find that approved highway segments (where people knew construction would start shortly, but work had not begun) had similar effects on Nazi support as highway segments already under construction.
We provide further support that propaganda and highway construction complemented each other – where radio signal strength was high and the new roads were under construction, pro-
10Since the election in 11/1933 and the referendum in 8/1934 are not directly comparable, we use the difference in standardized vote shares with mean zero and standard deviation one.
11 Our results reflect local differences in voting; for the country as a whole, effects may well have been larger because people did not only react to changes at the local level – but also because they saw progress in the country as a whole.
12 In other words, the Nazi propaganda successfully attributed economic performance to its economic policies – a process that is generally challenging because of economic volatility (Buera, Monge-Naranjo, and Primiceri 2011).
Nazi votes increased particularly strongly. In contrast, without radio coverage, the roads themselves had a negligible effect on voting behavior. One interpretation of these results is that Autobahn construction in combination with radio propaganda won “hearts and minds” of voters; another possibility is that the regime’s capacity to see through a major construction project also raised the expected cost of opposition.13 Which of these two channels dominated is not crucial for the main purpose of the referendum in 1934 – showcasing almost uniform popular support for Hitler, thereby signaling the regime’s popularity and stability (Evans 2006; Egorov and Sonin 2014).
Autobahn building also showcased the ability to get things done.14 Hitler announced the plans for new roads shortly after coming to power; within 9 months, he broke ground on the first stretch of motorway. Demonstrations of government competence were particularly attractive in areas where political turmoil had reigned. Weimar’s federal states with more unstable government in 1919-33 showed systematically larger vote-winning effects of highway construction.15 Regime propaganda also exploited the highways as powerful symbols of an energetic government overcoming “democratic gridlock” and the widely-lamented disorder of the Weimar Republic (Evans 2006).
The paper proceeds as follows. We review the related literature in Section 2, and then explain the historical background and context of motorway building in Section 3. Section 4 presents our data. In Section 5, we show the main empirical results. Section 6 presents instrumental variable results, and Section 7 examines channels through which the Nazi regime’s road building influenced voting. Section 8 demonstrates the robustness of our findings, and Section 9 concludes.
13 The latter explanation is less likely, given the regime’s wild popularity in later years. Raising the perceived cost of opposition would be a form of implicit intimidation. Explicit intimidation and fraud, on the other hand, are unlikely explanations for our finding, as we argue in Appendix A.4 (where we perform a number of “election forensics” tests and find no evidence for a relationship between Autobahn construction and fraud). Importantly, the Nazis were already in power during the November 1933 election. This makes it less likely that cross-sectional differences in intimidation or fraud accounted for electoral success; only a differential increase in (explicit) intimidation or fraud in areas with Autobahn construction could contaminate our results.
14 It also demonstrated an effective end of the austerity policies of the pre-1933 era that had been implemented by successive Weimar governments (Shand 1984).
15 Speeches by conservative politicians – and not only the Nazis – frequently referred to an alleged lack of speed and decisiveness in democratic decision-making. Indeed, even before Hitler’s seizure of power, plans for a new, authoritarian constitution were being proposed by many conservative politicians (cf. Stackelberg and Winkle 2013).
2 Related Literature
Our work contributes to research on the political economy of regime change (Acemoglu and Robinson 2000), elections and the entrenchment of dictatorships (Egorov and Sonin 2014;
Simpser 2013; Jessen and Richter 2011), the role of elections in autocracies (Gandhi and Lust- Okar 2009) and on interactions between the military and old elites (Finer 1976; Acemoglu, Ticchi, and Vindigni 2010). Closely linked is work on the origins of totalitarian dictatorships, much of which emphasizes differences between normal autocracies and regimes like the Nazi dictatorship or Communist rule in Russia. Theories of “mass society” focus on industrialization and the associated rise of a large group of economically marginal individuals who have lost their traditional roots (Ortega y Gasset 1993; Arendt 1973). These in turn are said to create a fertile recruiting ground for totalitarian ideology, from both the left and the right.16 Schmitt (1926), on the other hand, emphasized the need for an – alleged – external or internal threat for totalitarian states to consolidate.
Our research also relates to the rich literature on the electoral benefits of income transfers and infrastructure projects. There is evidence that politically motivated income transfers and federal spending can affect voting behavior (Manacorda 2011; Levitt and Snyder 1997), but aggregate patterns are often inconclusive (Stein and Bickers 1994). In a classic paper, Berman et al. (2011) examine under what conditions an occupying force can win the ‘hearts and minds’
of the occupied, and conclude that public service provision can have a decisive effect in reducing opposition. Also, public spending is often targeted at areas with a more informed electorate (Strömberg 2004).17 Recent research has also generated new insights into the economic effects of major infrastructure projects. While an early literature had concluded that the invention of the railways did not matter significantly for growth (Fogel 1964), there is now ample evidence that, for example, the building of India’s national railway network reduced transport costs, and increased trade (Donaldson forthcoming). Similarly, better access to transport infrastructure in China and Prussia boosted GDP (Banerjee et al. 2012; Hornung
16 Applications of this approach to the German context include Shirer (1960) and Stern (1972).
17 Along the same lines, Finan and Mazzocco (2014) show that politicians with greater electoral incentives transfer more resources to areas where they expect higher political returns. Larreguy et al. (2015) show how central-government policy interventions boost support for the federal incumbent while reducing the influence of local politicians.
2015)18 and increased land values (Donaldson and Hornbeck 2016). There are also well- documented effects on urban layout (Baum-Snow 2007), city growth (Duranton and Turner 2012) and skill premia in urban areas (Michaels 2008).
Relative to the existing literature, we make a number of contributions: First, we demonstrate the political benefits of infrastructure spending on electoral outcomes, helping to entrench the Nazi dictatorship. At a crucial moment, when the Hitler regime needed to showcase its popularity, Autobahn building boosted support. We thus contribute to the literature that studies regime change in general and the rise of the Nazis in Germany more specifically (King et al. 2008; Bracher 1978). Second, we provide new evidence of the channels through which infrastructure spending can make a difference. In the Nazi case, there were real economic benefits – but they are too small to account for the overall gains in electoral support. Instead, we show that propaganda can magnify the effects of locally successful policies, convincing voters far and wide that the new regime is getting the country out of its slump. Third, we offer suggestive evidence on the conditions under which elections can boost support for an autocratic regime – a key question in the literature on voting in non-democratic settings (Gandhi and Lust-Okar 2009). We find that road building was most effective in swaying voters in areas that had previously supported moderate parties, and in Catholic areas, which were typically skeptical of the Nazis. On the other hand, in areas with high support for the communists (such as worker strongholds), highways were less effective in garnering votes.
Lizzeri and Persico (2001) argue that in electoral regimes where the margin of victory matters, public goods are more likely to be provided, and pork barrel spending is lower. Our result on the Nazi regime building highways is related, but it goes further. Goebbels’ propaganda emphasized that roads are public goods. This generated important synergies with actual construction, enabling the regime to show near-universal support. In this sense, the Autobahn’s success in boosting pro-regime votes relied more on a perceived ‘competence’
channel (Rogoff 1990) than on any direct economic benefits. Thus, we provide evidence in support of arguments that autocracy may be attractive to voters because of its alleged ability to ‘get things done’ (Djankov et al. 2003).
18 In contrast, Faber (2014) finds adverse effects of highways on GDP growth in newly-connected peripheral counties in China.
3 Historical Background
In this section, we briefly describe motivations behind the building of the Autobahn network and its antecedents. We also discuss the nature of early Nazi elections and the growing strength of the regime.
3.a Motorway building under the Nazis
Road building became a government priority soon after the ‘seizure of power’. At the Berlin Motor Show – only 11 days after becoming Chancellor – Hitler presented a set of measures to boost car ownership and use, consisting of subsidies, road-building, and cheaper cars.19 By the summer of 1933, a new publicly-owned company had been founded to build and operate highways Germany-wide. Plans for the network built on work by a think tank, the STUFA (Vahrenkamp 2010). In some cases, the trajectory of the actual roads was decided by Hitler himself, who insisted on scenic routes (Todt 1934).
Construction began at many points simultaneously. Figure 2 shows the 1934 highway network.
Thick black segments were under construction; double-ruled segments were approved for construction, but not yet begun; and light grey lines indicate planned segments not yet approved for construction.20 Segments under construction, and trajectories approved for building, had high local visibility: Actual work was taking place, and the path of the new road was officially announced and staked out. Planned segments, in contrast, were only known to the planning officials. In 22 locations, construction was under way less than a year after the start of the project. None of the highway segments was actually open for traffic by the time of the plebiscite in August 1934.
Large-scale construction only began after the election in November 1933 – a fact that we exploit in our empirical analysis. Figure A.1 in the appendix shows employment in Autobahn
19 In the Rhineland, another – unrelated – project connected Bonn and Cologne. Konrad Adenauer, later Chancellor of the Federal Republic of Germany, coordinated the building in a bid to reduce unemployment. This first highway opened in 1932. At the time, Italy had already completed the first high-speed roads reserved for car traffic.
20 We digitized the September 1934 map from Todt (1934), which is the closest available to August 1934. The transition between highway segments “approved for construction” and “under construction” in Figure 2 is fluid, and even the historical maps are not completely clear about the exact timing when construction began. For example, a few smaller segments are listed as “under construction” in the May 1934 map, but as “approved for construction” in the November 1934 map. We use “under construction” as our main ‘treatment’ variable and document the robustness of results to including “approved for construction” in Section 7.e. Whenever we refer to “highways” in the empirics, we mean segments that were listed as “under construction.”
construction, by month, for the period 1933-34 (Humann 2011). Employment in November 1933 was 3,000 men, 5% of the level reached by August 1934. Earlier months saw even lower numbers of workers. From April 1934, at least 20,000 men were employed. In August, it reached 59,000. This was not yet the high water mark of Autobahn employment, but it was higher than in any preceding month, and equivalent to half of the all-time peak of employment (June 1936; 121,000 workers).
Together with rearmament, the Autobahn was a key part of Keynesian demand stimulus by the Hitler government. In line with the regime’s propaganda, many observers took it for granted that building the new highway network reduced unemployment. John Maynard Keynes himself, in the German edition of his General Theory, argued that Nazi spending policies after 1933 exemplified the superiority of totalitarian regimes in implementing the
“right” policies to overcome the slump.21 Quantitative research has since established that neither military spending nor highway construction were probably responsible for Germany’s recovery after 1933 (Ritschl 1998). Initially planned to employ up to 600,000 workers, motorway building never came close.22 Instead, the rapid rise in output under Hitler is typically explained by the strength of a cyclical upswing, helped by an end to deflation and declining uncertainty.
Nazi propaganda exploited the motorway from the beginning. The regime emphasized highway construction as an integral part of its war on unemployment (Arbeitsschlacht).23 Propaganda Minister Josef Goebbels ensured that work started simultaneously at many locations. This did not maximize use value, but made construction work visible across the country (Shand 1984). Hitler turned the first sod of earth in September 1933. In his opening speech, he argued that the new roads would bear witness to “our [the regime’s] devotion, our diligence, our ability, and our decisiveness” (Schütz and Gruber 1996). In the first month of the newly-founded Autobahn company’s existence, the Völkischer Beobachter – the leading Nazi Party paper – made construction progress front-page news no fewer than four times.
21 Keynes (1936). Scholars from Karl Schiller (1936) to Richard Overy (1975) argued along similar lines.
22 At its peak, only 121,000 Germans were working in highway construction (Humann 2011). This should be compared with a decline in unemployment from 6 million in January 1933 to 2.5 million in the summer of 1934.
23 Literally, “battle for labor.“
Radio similarly played a prominent role – the start of construction was broadcast live to millions of listeners, including speeches by Hitler and Goebbels.
The regime celebrated the opening of each new stretch of motorway. The first segment near Frankfurt was opened by Hitler himself in May 1935, with 90,000 supporters lining the road.
By 1936, some 1,000 km of road (out of 9,000 planned) had been finished. Each opening was covered on the radio, in the press, and by the cinema news reels (Schütz 1995).24
Why was highway building given priority, instead of other public works programs? Road building as a make-work measure had been discussed extensively during the Great Depression, but no large-scale construction had taken place. The actual building of the highways signaled a regime change – a break with austerity (Ritschl 2003). Party propaganda told readers and listeners that the new highways proved that Germany was booming thanks to the Nazi regime.
Ever since, the Autobahn project had symbolic character: “Hitler breaking new ground … – the picture became an icon of the year immediately after 1933, a symbol for everything Autobahn construction seemed to stand for: energy, directness and dynamism of the national socialist movement…” (Schütz and Gruber 1996, p.43).
Motorway workers themselves were often skeptical of the Nazi regime. They came from among the unskilled and unemployed in big cities. Workers typically lived in barracks, with harsh discipline and low wages. Sometimes, disaffected workers even painted anti-Nazi slogans on construction lorries (Evans 2006), and went on strike
Germany’s car ownership rate in 1933 was low – approximately one quarter of England’s or France’s. Most transport of goods and people took place via rail. The new regime intended to boost the German car industry by all means possible, and not simply via road-building. Hitler had high hopes for the automobile industry, for employment and war production. A tax exemption on car purchases boosted car production. Between 1932 and 1938, the total number of cars, motorcycles, and trucks on German roads doubled (Evans 2006).
There were few military advantages to road-building. Almost all troop and supply movements before and during World War II were by rail. Nonetheless, the growth of the motorcar industry expanded war manufacturing capacity. Boosting the mobility of army units was an aim of
24 In addition, the Autobahn was also celebrated as an aesthetic innovation. The Autobahn company commissioned a number of artists to produce paintings of road segments, bridges, ramps, and construction work.
A book containing reproductions of these paintings sold over 50,000 copies (Vahrenkamp 2010).
most armed forces after 1920 (van Creveld 1977). Increasing car ownership and the number of trucks in Germany benefited the army because private vehicles could be confiscated in wartime -- the invasion of France in 1940 used some 15,000 trucks requisitioned from private industry (Vahrenkamp 2010).
3.b 1933 Elections and the 1934 Plebiscite
Figure 3 illustrates the timeline of elections and highway building. When Germans went to the polls in March 1933, the Hitler government had already been in power for over a month.
Nonetheless, elections were still relatively fair, with limited intimidation at the polls. The Communist Party had been banned, but all other parties that had competed during the last free election in November 1932 were still on the ballot in March 1933. Despite a massive propaganda campaign (Adena et al. 2015), the NSDAP failed to win an absolute majority, receiving 44 percent of the vote.
In November 1933, the regime held fresh elections. Over the summer, all parties except the NSDAP had been banned. On average, the Nazi Party won 92 percent of the popular vote, more than doubling its vote share from March.25 Voting in November 1933 was not free and fair; storm troopers stood guard at the voting booths. Citizens were strongly “encouraged” to vote publicly to demonstrate their support of the Nazi regime. Evans (2006) observes that
“Intimidation was particularly evident during the national plebiscites and elections
… Under the Third Reich, plebiscites and elections became propaganda exercises
… to provide the appearance of popular legitimacy for controversial measures.”
Nonetheless, opposition was not zero. On average, eight percent of all Germans effectively voted against the Nazi list (by spoiling their ballot papers – voting “no” was not possible in Nov. 1933). In some areas, there was massive opposition – in the old Hanseatic city of Lübeck, for example, 40,824 voters did not vote “yes” for the NSDAP list, out of 111,911 votes cast – a proportion of 36.5 percent. Hamburg and Berlin also registered high levels of dissent, with 27 and 26 percent of voters refusing to support the Nazi list, respectively. In other cities, less than 1% of votes were spoiled.26
25 In parallel with the parliamentary election, voters were also asked to approve Germany’s leaving the League of Nations. This proposal was wildly popular since the League of Nations was closely associated in the minds of Germans with the (hated) Versailles settlement that saddled Germany with massive reparations (Evans 2006).
This referendum received 95% support.
26 There are also several smaller towns where support reached 100%.
In August 1934, Germans were called to the polls again to vote on the Plebiscite that would make Hitler both Chancellor and President. While overall support remained high, it declined slightly on average compared with November 33 – 89.9% voted with yes in August 1934.27 3.c Crisis and Entrenchment of the Nazi Dictatorship 1933-34
Once in office, the Nazi leadership quickly asserted control over the police; violence against opponents – suspected or real – was frequent (Evans 2006). Despite this ruthlessness, the regime was not firmly established during its first 18 months. In 1934, opposition to the regime increased: The most radical part of the Nazi movement, the storm troopers (SA) threatened a
“second revolution.”28 Middle class voters, often supporters of the NSDAP before 1933, feared wider chaos (Behnken and Rinner 1980):
“The moment was … critical for the regime. … enthusiasm of the ‘national revolution’ in 1933 had discernibly fallen off ... The brownshirts were not the only section of the population to feel disappointed ... Social Democratic agents reported
… that people were apathetic, constantly complaining, and telling endless political jokes about the Nazi leaders. Nazi meetings were poorly attended … The educated classes feared that the disorder caused by the stormtroopers might spill over into chaos or, worse, Bolshevism.” (Evans 2006)
By 1934, the Nazi leadership feared that the conservative members of the government might join forces with the army and overthrow the Hitler regime (Evans 2006). Paul von Hindenburg was still President, close friends with Franz von Papen, a former Chancellor now serving as Vice Chancellor. In June 1934, von Papen gave a famous speech in Marburg in which he decried violence and lawlessness by the SA, and condemned the personality cult of Hitler.
Thereafter, his public appearances were often greeted with the shout “Heil Marburg.” Time Magazine concluded that Hitler was not a “real dictator.”29 At this moment, the Defense Minister, General Werner von Blomberg, threatened Hitler with a coup by the army if the SA
27 While the November 1933 election (for the Nazi Party) and the 1934 referendum (for Hitler in person) are clearly distinct, there is no obvious downward bias – right down to the end in 1945, Hitler personally was much more popular than the Nazi Party.
28 The SA grew out of street-fighting paramilitaries; its leaders envisioned themselves as a Nazi People’s Army, and many pursued dreams of a far more left-wing agenda including wholesale nationalization of many industries.
Threats to Hitler’s leadership, however, were largely invented to justify the crackdown on the SA in the summer of 1934 (Evans 2006).
29 “…if Adolf Hitler came home with a swelled head and hot new ideas for dictatorship from his visit to Benito Mussolini, certainly last week he was dexterously chilled and shrunk...” "Germany: Second Revolution", Time, 2.July 1934.
was not brought to heel (Wheeler-Bennett 1964). Eventually, Hitler had both the leadership of the SA and leading conservatives murdered, claiming that the victims had been plotting a putsch.
The conflicts and threats of the summer of 1934 show that the Nazi regime was still far from its later, omnipotent position; popular support could by no means be taken for granted. Indeed, knowledgeable observers concluded that there was “evident shakiness in high Nazi places”
and that “Adolf Hitler [got] the scare of his career” (Time July 2, 1934). An opportunity to showcase overwhelming popular support was of key importance to the regime at this juncture.
It was only after Hitler became both Chancellor and President as a result of the 1934 Plebiscite, where a large share of the population publicly supported the Führer’s new powers, that the regime became deeply entrenched.
3.d Political Instability
Germany’s first attempt at democracy, the Weimar Republic, was synonymous with political instability; governments changed with alarming frequency. The perceived “chaos” of democratic governance was often contrasted with order before 1918; instability was one key reason why support for democratic rule waned. As the novelist Stefan Zweig (1942) observed:
“… inflation, unemployment, [and] the political crises had disturbed the German people deeply; an incredible desire for order spread amongst the German people... And whoever promised order … immediately had hundreds of thousands behind him.”
Despite great turmoil in the country as a whole, some federal states showed relative stability, with the same prime ministers and parties in power for long stretches of time. Prussia is one notable example (Orlow 1986).
4 Data
We use voting records for more than 3,000 towns and cities, covering the entire area of Weimar Germany (Statistisches Reichsamt 1934), combined with demographic and socio- economic information from the 1925 and 1933 censuses (Falter and Hänisch 1990). To this, we add newly coded geographical information from historical maps of the (planned and built) German motorway network, construction of which began after the summer of 1933, as well as data on vehicle ownership, radio signal strength, and the political stability of Weimar states.
4.a Highway Construction and Balancedness
The general plan for the Autobahn envisaged connections between the largest cities (Figure 2). Of the 3,231 towns and cities in our sample, 1,982 were within 20 km of the planned Autobahn. A little more than a third (1,249) were further away (Table 1). Out of the 1,982 locations close to the planned network, 1,075 saw actual construction by the summer of 1934 – 54% of the planned total.
Some socio-economic characteristics differed between towns close to the highway network and more distant ones. Table 2 gives an overview, showing the sample mean of several socio- economic variables from the 1925 and 1933 German censuses in column 1, the average for cities within 20 km of the highway network (planned or built) in column 2, and the means for cities with and without actual highway construction, among those near the planned network (columns 3 and 4). Cities near the planned highway network were more populous than the rest; initial unemployment (in 1933), the blue-collar share, and industrial employment were also somewhat higher, while there were fewer Catholics than in the sample overall. The share of Jewish population was the same. Comparing columns 3 and 4 shows that construction began in those parts of the planned network that were closer to larger, more industrial cities, and in more Protestant areas. This gives rise to endogeneity concerns, because support for the Nazis also varied with socio-economic factors. In our empirical analysis we address this issue in a variety of ways, by adding explicit controls and city fixed effects, entropy weighting to create a balanced sample, and the use of least-cost-paths as an instrument for actual highway location. Importantly, pre-existing support for the Nazi regime did not affect systematically the location of highway construction, as shown by the balanced NSDAP vote share in March 1933 in Table 2. We explore this in more detail below, showing that there were also no differential pre-trends in Nazi support before highway construction started.
4.b Elections and Plebiscites
Our main analysis focuses on the change in the share of voters supporting the Nazi regime between the November 1933 election and the 1934 plebiscite.As a proxy for underlying Nazi support, we also use the NSDAP vote share in the March 1933 election – after Hitler had been appointed as Chancellor, but when other parties were still permitted at the polls. Figure A.2 in the appendix plots the share of “pro-Nazi” votes in the three elections we focus on. Since elections after March 1933 were no longer fair and free, the officially registered support for
the regime at the polls surged until November 1933. Between November 1933 and August 1934, the share of pro-Nazi votes declined somewhat – if we want to disregard the fact that the nature of the vote changed, too.
To make the different elections comparable, we rescale vote shares in our empirical analysis, transforming electoral ‘pro-Nazi’ votes for each election into a standardized variable with zero mean and unit standard deviation. Our main outcome variable is NSbroad, which is defined as the share of yes votes relative to all eligible voters. This variable counts nonvoters as opposition to the Nazi regime – which in many cases is justified given the high pressure for turnout (see Section 3). In Appendix A.6, we show that our results also hold when using a narrow measure (NSnarr), which is defined as the share of yes-votes relative to actual voters;
it is thus unaffected by voter turnout (and by potential unobserved spatial variation in the pressure to vote).30
4.c Radio
From the 1920s onwards, Germany had a highly developed, government-owned system of radio stations (Bausch 1956). By the 1930s, governments regularly used radio programs to bolster support (Adena et al. 2015). There is detailed data on the number of radio subscribers in various parts of Germany, and on the strength of radio signals. Since the purchase of a radio subscription may itself be a function of political preferences, we follow Adena et al. (2015) and focus on city-specific signal strength, as determined by the power and location of transmitters interacting with terrain characteristics.31 We find that listenership increased with signal strength in a flexible non-parametric estimation (see Appendix A.7 for detail).
4.d Political Instability of Weimar States
To measure political instability at the state level, we follow Satyanath et al. (2017), who measure political stability as the first principle component of three indicators: the percentage of time that i) the longest-serving state government was in office, ii) the longest-serving party
30 Note that both measures count invalid votes as opposition to the Nazi regime. In fact, the November 1933 election did not allow for a “no” vote, or for votes for any other parties. Thus, conditional on voting, invalidating the ballot was the only way for voters to express their discontent with the Nazi regime. The 1934 referendum, in contrast, included an option to vote “no.” This is another reason why the two elections are not directly comparable, motivating our use of standardized vote shares, rather than comparing levels.
31 We use predictions from the irregular terrain model. Ruben Enikolopov kindly provided us with city-specific signal strength data for all locations in our dataset from the implementation of the radio diffusion model in Adena et al. (2015).
was in office (possibly in different coalitions), and iii) whether a state was governed by at least one party from the “Weimar coalition.”In combination, these three variables capture political turmoil at the federal state level during the Weimar period.32
5 Main Empirical Results
In this section, we show that support for the Nazi regime increased significantly more where the new motorways were being built. Before presenting econometric estimates, we first illustrate our main finding. Figure 4 maps changes in support for the Nazi regime between November 1933 and August 1934.33 The darker the red on the map, the greater the (residual) electoral gains of the Nazi Party. Solid black lines are roads under construction; dashed ones, roads approved but not yet being built. On average, areas through which the new highways passed saw much greater gains in support for the Nazis than the rest. This is particularly true in East Prussia, the North of Germany, in the West around the Ruhr, and in the area around Frankfurt. While there are areas with significant increases in support without road-building (such as along the shoreline of the North Sea near Holland), they are relatively rare.
5.a Baseline results: Change in Nazi support, November ’33 – August ‘34
In what follows, we present our baseline specifications, estimating regressions of the form:
∆NSi = α + βDi + γXi + εi (1)
where ∆NSi is the change in (standardized) pro-Nazi votes between November 1933 and August 1934 in city i, Di is its distance to the nearest highway segment under construction, Xi
is a vector of controls, α is a constant, and εi is the error term. If Di was randomly assigned, β would reflect the causal effect of motorway building on support for the Nazi regime. We present OLS results first, and then discuss potential challenges to identification, followed by IV results.
In Table 3, Panel A, we first show the simplest specification without controls in column 1.
We find a negative and highly significant coefficient on distance to highways. To gauge its magnitude, we calculate the implied vote change (in percentage points) when going from one
32 Satyanath et al. (2017) end their coding period before the Prussian coup d’état in July 1932, which is often considered the beginning of the end of the Weimar democracy. We use their original measure, but have also extended their coding period to January 1933 as a robustness check.
33 We plot effects after accounting for log city population and unemployment in 1933, as well as regional fixed effects corresponding to 77 administrative districts in Weimar Germany (Regierungsbezirke).
to 100 km distance to highway construction.34 This yields a vote gain of 2.9 p.p. In column 2, we add our baseline controls as well as the Nazi Party vote share in the preceding November 1933 election. The coefficient on highways declines but remains highly significant, and it rises again when we add fixed effects for 77 administrative districts in column 3. Adding the latter means that we exploit only the distance to highway segments within each district, differencing out any regionally-based shifts in voting patterns. Our results in column 3 thus imply that, relative to all the other towns in the same district, those closest to the new highways saw particularly large gains in Nazi support.
In terms of control variables, the negative coefficient on initial pro-Nazi votes in November 1933 is significant and negative, which reflects a mechanical effect – places with close-to- 100% support could hardly gain additional votes. The coefficient on city population size is negative and significant – more populous places saw less of an increase in Nazi support.
Finally, the coefficient on unemployment in 1933 is ambiguous, switching signs and becoming insignificant when we add district fixed effects.
In column 4, we add additional socio-economic controls – the share of blue-collar workers, of Jews, of Catholics, and of industrial workers – the significance of the distance-to-highway variable is not affected, but it declines somewhat in size. Finally, we define a dichotomous variable that takes on value one for towns or cities that were within 20 km of highways under construction, and zero otherwise. In columns 5 and 6, we use this alternative variable to repeat the specifications from columns 2 and 3.35 We again find highly significant coefficients that suggest an increase in support by 0.12 standard deviations (corresponding to an increase in pro-Nazi votes by about one percentage point) if a town was close to the Autobahn. Appendix A.2 shows that alternative cut-offs for distance to highways lead to very similar results.
34 To obtain vote differences in levels for Nov’33-Aug’34, we multiply the result for standardized vote differences by the standard deviation of vote differences in levels.
35 Note that these regressions essentially reflect a difference-in-difference specification, where `treatment’ is a dummy for highway proximity. Thus, treatment effects are estimated as changes in pro-Nazi votes before vs.
after highway construction, in areas near vs. distant from highway segments. The full diff-in-diff specification (running regressions with Nazi votes in levels for Nov’33 and Aug’34, with city and year fixed effects) yields almost identical results (available upon request).
Panel B of Table 3 we allow for arbitrary spatial dependence across observations, following Colella et al. (2019).36 The standard errors for the coefficients on distance to highway under construction are very similar to our main results in Panel A of Table 3. This suggests that our results are not confounded by spatial correlation.
For completeness, Panel C of Table 3 presents results using unstandardized changes in pro- Nazi votes as dependent variable. This specification relies on the (debatable) assumption that vote shares in the November 1933 election and in the August 1934 referendum are comparable.
Nevertheless, we find very similar effects of highway construction on Nazi support.
5.b Persuasion rates
How effective were highways in winning voters for the Nazi government? And how large were electoral gains overall? We gauge the effectiveness of highway building by looking at persuasion rates – the extent to which exposure to the highway was associated with voters switching from opposition to support of the government (Della Vigna and Gentzkow 2010).
Panel C of Table 3 reports our estimates of persuasion rates – the share of voters who changed their minds because of highway construction. To calculate these rates, we assume that, in the log-linear specification (cols 1-4), voters in areas further than 100 km from the highway were not affected, thus forming our ‘control’ group. With this, using the method in Della Vigna and Gentzkow (2010), we calculate the increase in Nazi support when going from 100 km of distance to highway construction to 1 km, relative to the share of voters who were left to be persuaded to support the Nazi regime. For the dummy specifications in cols 5 and 6, we assume that treatment is zero beyond 20 km distance. We find substantial persuasion rates, ranging from of 8 to 17% for the log-linear specification, and 6% in the dummy specification.37 Note that these figures are a lower bound; nationwide increases in Nazi support due to highways are not captured by the persuasion rates.
36 We consider all cities with distance less than 700 km spatially contiguous and assigned them a nonzero spatial weight. The cutoff of 700 km reflects the average distance of the most important East-West (Berlin-Aachen) and North-South (Hamburg-Munich) highway connections planned by the Nazi regime.
37 For comparison – the voting studies surveyed by Della Vigna and Gentzkow (2010) show a range of persuasion rates in elections of 2 to 20%, with an average of 11.9%.
5.c Voting patterns before and after highway construction
Are our baseline findings in Table 3 specific to the period November 1933 to August 1934, when road building got under way in earnest – or do they reflect pre-existing differences or differential trends?
To examine this question, we first add the last relatively free election of March 1933 to our analysis. We find that votes for the Nazi Party in March 1933 were not significantly correlated with distance to highways that would be built from late 1933 onwards (see Table A.7 in the appendix for detail). Next, we examine the change in Nazi support before and after highway construction began on a large scale. This striking difference in effects is illustrated in Figure 5.38 The left and middle panels serve as placebos, illustrating the change in pro-Nazi votes between September 1930 and March 1933 (before highway construction), as well as between March 1933 and November 1933, when only very little highway construction had taken place.
There is essentially no relationship between distance to highways and change in Nazi support.
This pattern changes dramatically after November 1933, when highway building took off:
The right panel of Figure 5 shows that by August 1934, it was the areas closest to the highway that saw the biggest relative gains in Nazi support. The difference between the coefficients in March/November 1933 and August 1934 is crucial for our argument. It implies that distance from the highway only becomes a predictor of Nazi support after construction began on a large scale – after November 1933.
Next, we generalize the analysis by using panel regressions. We estimate the relationship NSit = αi + δt + βDi×δt + γXi×δt + εit (2) where NSit are pro-Nazi votes in city i in election t, Di is city i’s distance from the nearest highway segment under construction, Xi is a vector of city-level controls, αi and δt are city and election fixed effects, and εit is the error term. Note that we interact Di and all controls with year dummies. This allows the coefficients on distance to highways and controls to vary in each period, thereby also effectively running placebo tests for the periods before road construction began.
We present our panel results in Table 4. In columns 1-4, we pool election data on the success
38 Given that regular scatterplots with every data point would become too crowded for visual interpretation, we use binscatter plots, grouping the x-axis into 25 equal-sized bins. To allow for a more immediate interpretation of the x-axis, we use distance in km, rather than log-km. Results are very similar when we use logs instead.
of the Nazi Party from the early years of dictatorship (Mar. 1933- Aug. ‘34). We find a negative and significant coefficient on distance to highway construction only for the August 1934 election; for all earlier elections, the interaction with the highway distance variable reveals no statistically significant or economically meaningful relationship. The non-results for March and November 1933 also imply that Autobahn construction was not used to reward districts with strong previous support for the Nazis; in other words, `favoritism’ in the sense of Burgess et al.’s (2015) finding for Kenya is probably not present in our data. These results are robust and hold when we interact our baseline controls (population and unemployment) with year dummies (col 2), when adding lagged Nazi Party votes (col 3), and when we add interactions of additional socio-economic controls with the year dummies (col 4).
In the last two columns in Table 4, we use data from all elections with city level data during the period 1924-34.39 We estimate both with fixed effects only (col 5), and with extended controls and lagged Nazi votes (col 6). Again, the 1934 referendum is the only period that shows a statistically significant relationship between Nazi support and distance to highway construction.
Overall, there is no evidence that Nazi support was either high or already growing in places where highways were (later) built (see also Table A.7 in the appendix). Instead, the entire effect of highway construction on electoral outcomes appears suddenly, and only for the period November 1933–August 1934.40
6 Identification
Given the highways’ propaganda value, their trajectories may have been chosen for political reasons. In this section, we focus on the period Nov. 1933 to Aug. 1934 and instrument the path of actual highway construction between given city pairs with terrain characteristics that facilitated road building.
39 The NSDAP was banned from the 1924 election as a result of the failed Beerhall putsch. Members of the banned NSDAP reconstituted themselves as a party under the label NSFP, which put forward joint lists with the DVFP. The DVFP absorbed much of the Nazi vote in the May 1924 election (Striesow 1981), and we use its standardized vote share in the panel in 1924.
40 In Appendix A.2 we perform an additional analysis, restricting the sample to towns within 20 km of planned highway construction. This increases the similarity of towns and cities in our sample, excluding those that were never to be connected to the highway system. Even within this narrowly defined subsample, we find a strong relationship between actual construction and Nazi support.
6.a IV-Results: Least cost paths
Our results could be affected by endogeneity bias if the Nazis targeted areas that were more likely to increase their support for the regime even in the absence of highway construction.
The Nazis could also have planned and built highways to reward (newly) loyal districts, or strong local Nazi officials may have been more successful at both attracting the highway and swaying voters. On the other hand, OLS results could also be downward biased, if Nazi highway building targeted areas where it was more difficult to win new supporters.
Endogeneity concerns cannot be dismissed out of hand – for example, Hitler himself intervened in the planning of the road from Munich to Salzburg, although largely on aesthetic grounds (Vahrenkamp 2010).
To address possible endogeneity concerns, we instrument for actual highway building with least-costs paths. Road construction cost is highly sensitive to the slope of the traversed terrain.
We use the Cost Path tool in ArcGIS to calculate the cheapest way to connect city pairs that appear in official German publications as terminal cities that were to be connected in the first wave of highway construction.41 Figure 6 plots least-cost paths (LCPs) and actual highway construction that began before August 1934. They overlap to a large extent. Even where the LCP does not coincide exactly with the actual motorway trajectory, differences are typically small. The only larger deviations are in North Germany, where the terrain is generally flat and small differences in cost can lead to quite different paths.
Least cost paths have substantial explanatory power for actual highway construction: Out of the 3,276 towns and cities in our sample, about one-half (1,602) lie within 20 km of a least cost path. Of these, 1,404 (87.6%) also lie within 20 km of the actually planned highway network, and 914 (57.1%) of them saw actual construction activity by the summer of 1934.
In contrast, of the 1,674 towns and cities that were more than 20 km away from least-cost paths, only 183 (10.9%) saw construction.
41 We compute least-cost paths for all 38 city-pair connections listed in Jahnke (1936). See Appendix A.1 for details. Related work using geographical characteristics or earlier transport infrastructure for identification includes Baum-Snow (2007), Donaldson and Hornbeck (2013), Banerjee et al. (2012), and Faber (2014). We do not use the network analysis as implemented by Faber (2014), for example, who uses Kruskal’s minimum spanning tree algorithm to pin down a cost minimizing network structure. As Figure 2 confirms, the Nazi building of the Autobahn did not follow a network logic, with an increasing set of cities connected to existing roads. Instead, the regime initially connected city pairs, and it started to build in multiple disconnected locations all over the country – delaying the opening of the first useable road, but making the project more visible.
Our instrumental variable is the distance of each city from the least cost paths (LCPs).
Crucially, all regressions exclude the 27 terminal cities, i.e., the end points between which LCPs are computed.42 Before presenting our IV results, we briefly discuss their interpretation.
Importantly, least cost paths affect the planning of highways, while the electoral effects we are interested in are due to actual construction. Planning of highways translated into highway construction in some districts by 1934 – depending on the timing of construction. Our IV strategy estimates the average effect of highway construction on pro-Nazi votes for those cities whose ‘treatment status’ (proximity to highway construction) was affected by the instrument (proximity to LCPs). Using common IV terminology, we estimate the average treatment effect for “compliers”(cities where proximity to LCPs did results in construction).
In contrast, cities close to LCPs where no construction occurred by 1934 (“never-takers”) do not affect our estimate.
Table 5 presents our IV results. We first show results for the reduced form, regressing change in support for the Nazi Party on distance to LCPs. We find strong and significant negative coefficients, both without controls (col 1) and with the full set of controls (col 2). In column 3, we perform a placebo analysis for vote gains over the period March-November 1933, showing that distance to LCPs does not predict Nazi vote gains before highway construction began. Next, we return to our main period of analysis and demonstrate the strength of our instrument (cols 3 and 4). The first stage is powerful, with F-statistics above 500. Finally we present the IV results in columns 5 and 6. We find highly significant coefficients on instrumented distance to highways in the second stage, with comparable – albeit larger – magnitude than our OLS estimates in Table 3.43
7 Channels
Why did the Autobahn succeed in raising electoral support for the Nazis? To gain insight into possible mechanisms, we first look at one direct economic benefit of highway construction –
42 Our baseline OLS results (Table 3, Panel A, col 2) are identical when we exclude the terminal cities.
43 When all controls and district fixed effects are included (col 6), the IV coefficient is almost twice as large as its OLS counterpart in the same specification (col 4 in Table 3, Panel A). This may be due to measurement error in the actual location of highway construction (see footnote 20). In Appendix A.2 we perform a more restrictive IV analysis, using only connections where both terminal cities belonged to the top-20 in terms of population in 1933 – i.e., cities that most sensible road planners would have planned to connect. We find strong and highly significant results that closely resemble those in Table 5. This makes it unlikely that the Nazi regime strategically picked terminal cities in order to ‘treat’ the areas in between.
unemployment reduction. We then analyze other explanations; in particular, whether road building ‘worked’ because it served as a signal of government competence.
7.a Unemployment
As Hitler came to power, unemployment was by far the most prominent economic policy issue.
Goebbels’ propaganda told everyone with eyes to read and ears to hear that the country’s recovery after January 1933 reflected successful Nazi policies – and no intervention was more visible throughout the country (or more talked about in the propaganda) than highway construction (Todt 1934).
Can vote gains for the Nazis be explained by employment creation due to highway construction? To answer this question, we collect data on unemployment for all 253 cities where this information is available in February 1935 (for details on data construction see Appendix A.3).44 This allows us to compute the change in the unemployment rate between June 1933 and February 1935 – the period with available data that is closest to our main period of interest between the November 1933 election and the referendum in August 1934. We first check whether changes in local unemployment were related to Nazi support.45 The left panel of Figure 7 shows that this was the case: Gains in Nazi support were greatest where unemployment fell the most. The corresponding coefficient implies that a one p.p. reduction in unemployment was associated with a 0.026 std (equivalent to 0.11 p.p.) increase in Nazi support (see Appendix A.3 for regression results). This corresponds to an increase in Nazi support by 0.48 p.p. for a one-std drop in unemployment.
Was the decline in local unemployment, in turn, driven by highway construction? If this was the case, (un)employment would be a likely channel through which Autobahn construction boosted Nazi support. The right panel of Figure 7 shows that this is not the case. There is no relationship between the change in unemployment and distance to highways (controlling for initial unemployment). The corresponding coefficient is small, negative, and within a tight
44 We combine data on unemployment from the June 1933 census (which is available at the municipality level).
The 253 cities for which unemployment figures are available in February 1935 are relatively large, accounting for 23% of overall unemployment in the 1933 census. On average in this subsample, the unemployment rate fell by 12.2 percentage points – from 23.0% in June 1933 to 10.8% in February 1935. However, caution is warranted when comparing unemployment rates over time, since those in 1933 are from the Census, while the numbers in 1935 are derived from unemployment claims filed with the Labor Ministry (see Appendix A.3).
45 For consistency, we account for our baseline controls in Figure 7 and in the underlying regressions presented in Appendix A.3. The baseline controls are log city population and the initial unemployment rate in 1933, as well as the share of pro-Nazi votes in November 1933.
confidence interval around zero.46 Similar results hold in alternative specifications (see Table A.4 in the appendix). While we can only show this (non-)result in the subsample of 253 cities, it is nevertheless suggestive: Localjob-generation is unlikely to explain the effect of highways on Nazi support. In other words, pure economic voting based on improvements in the local labor market can probably not account for a sizeable part of the swing towards the Nazi government.
7.b Did Vehicle Ownership Matter for the Effect of Highways?
Direct effects could also come through vehicle ownership and the greater use-value of automobiles. Germany as a whole had quite low vehicle ownership, with only 674,000 cars on the road (including buses) in 1934, plus another 984,000 motorcycles – equivalent to 10 cars and 15 motorcycles per 1,000. Any benefits from using these vehicles would have had to be anticipated in August 1934, since new roads only opened from 1935 onwards. In Table 6, we stratify the sample by the density of motor vehicles (including buses, motorcycles, and cars, available at the province level from Frik 2004). There is no evidence of greater electoral gains in areas with higher vehicle ownership; the coefficients on distance to highway are very similar and (marginally) significant in both subsamples. Our results thus suggest that direct economic benefits are unlikely to account for the effect of highway construction on local Nazi support.
7.c Radio Propaganda and “Competence”
Next, we examine another hypothesis – that highway building convinced people of the regime’s competence, and of having the population’s best interests at heart. As Hitler came to power, unemployment was by far the most burning economic issue. Goebbels’ propaganda loudly proclaimed that the country’s recovery after January 1933 was a direct result of the regime’s policies – and none was more visible throughout the country (or more talked about in government propaganda) than highway construction. This suggests that places with greater exposure to propaganda should have experienced greater vote gains for the Nazis – and
46 This non-result echoes earlier arguments that doubted a role of highway construction in solving Germany’s aggregate unemployment problem (Ritschl 2003). Construction itself was often performed by brigades that moved along with highway segments (Evans 2006), making direct local employment effects unlikely.
Nevertheless, spending by construction workers or the placing of orders with local firms may have created employment around highway construction. In addition, local tourism may have benefited: Construction sites often became a popular destination for weekend trips (Eichner-Ramm 2008). Our empirical results suggest that such indirect job generation via highway construction was probably limited.
